Connect an AI client to the TSE API

A remote Model Context Protocol server covering all 173 TSE endpoints. It holds no credentials — you pass your own bearer token on each call — and stores nothing between requests.

Claude Code

claude mcp add --transport http tse https://mcp.travnet.app/mcp

Claude Desktop / any MCP client

{
  "mcpServers": {
    "tse": { "type": "http", "url": "https://mcp.travnet.app/mcp" }
  }
}

Tools

Tool	What it does
api_conventions	Auth, minor-unit money, idempotency, async search, hosts
list_areas	The map: every surface and area with counts
search_endpoints	Find endpoints by keyword, area or method
describe_endpoint	Full contract: every field, type, required flag and enum
call_endpoint	Execute a call against test or production

Five tools rather than 173. A tool list that large crowds out the model's context and makes selection worse; navigating the surface works better.

Safety

• Test by default. Production must be named explicitly.
• Writes need confirm=true. No POST, PATCH, PUT or DELETE happens by accident.
• Idempotency enforced. Endpoints that require an Idempotency-Key are refused without one.
• Your token, your call. Nothing is stored server-side.

Also available

• API console — browse and call all 173 endpoints
• llms-full.txt — full reference for AI clients
• openapi.json — OpenAPI 3.1 schema

© Sitenet Tech.